[wordup] Computer to pass turing test by 2029?

[Adam Shand]

Via: Mark Stevens <marks at verio.net>
Link: http://www.computerworld.com/managementtopics/management/opinion/story/0,10801,74084,00.html

Techno-entrepreneur Ray Kurzweil recently bet Lotus founder Mitch Kapor
$10,000 that a computer will pass the Turing test before 2029. British
computer science pioneer Alan Turing in 1950 said that if a human
interrogator, communicating blindly via text messages, couldn't
distinguish responses from a human from those of a computer, then the
computer could be deemed to have human intelligence. Kurzweil maintains
that by 2029, we will use nano-scale brain-scanning technology to
completely map and understand how the brain works and then
reverse-engineer it in a computer.

But Kapor says Kurzweil is making a dubious assumption - that the brain in
fact works like a computer, albeit a very complex one. Kapor argues that
we shouldn't engage in "distant extrapolation" of the brain-as-computer
metaphor. An overreliance on biological metaphors has been the undoing of
much of artificial intelligence, he says.

Indeed, we have used concepts from biology as computational metaphors ever
since Aetna Insurance installed its first "electronic brain" (an IBM 650)
in 1954. Such metaphors can give the layman a shallow inkling of what's
going on. But computer scientists and application developers would never
rely on them to guide their work, would they?

It turns out researchers are increasingly doing just that. Stephanie
Forrest, a computer scientist at the University of New Mexico, is building
systems that can detect hacker intrusions by imitating the human immune
system. A key challenge in computer security is determining what is normal
behavior and what is potentially harmful behavior in a computer or
network, especially when threats are changing regularly.

Forrest's systems automatically "discover" what is normal and what is not,
just as our immune systems have learned to do. Her software is largely
self-maintaining and doesn't require updating by experts. A computer
scientist at Los Alamos National Laboratory, appointed to an antiterrorism
research task force after Sept. 11, told me the technique holds great
promise for homeland security.

Now consider the ant. Rather than relying on complex, centralized logic,
systems that mimic ant behavior use many small, autonomous software
agents. With each acting on the simplest of rules, just as ants do, these
agents together can solve problems that, viewed as a whole, are enormously
complex. Today, software based on ant behavior is used for optimization
applications such as factory scheduling, vehicle routing and
telecommunications switching.

Meanwhile, other researchers are developing systems based on "evolutionary
computing" to solve factory scheduling and optimization problems. The
systems iterate through many trial solutions, breeding better and better
ones from the most promising parents in each generation of trials.
Solutions literally evolve in a process that selects the "fittest" in
every generation.

These biology-inspired algorithms aren't just the stuff of Ph.D.
dissertations. Researchers have really studied the immune system, ants,
evolution and other biological phenomena and have invented useful new
computational techniques based on them.

If you are a creator of software, or even a user of it, you might do well
to look to nontraditional sources, biological and otherwise, for
inspiration. And you might consider that the falling cost of computer
cycles makes problem-solving approaches that were only theoretically
possible a few years ago practical today.

Richard Gabriel, a computer scientist at Sun Microsystems, says he does
90% of his research on the Internet and that it can already pass the
Turing test. "That's a heck of a smart thing out there," he says. "I just
typed into Google, 'How do I change a tire?' and I got the answer on the
first page."

Even Kapor would have to admit that the Internet works a lot like the
brain. It consists of millions of loosely connected nodes (neurons) whose
relationships (synapses) change constantly. Nobody has trained the
Internet to answer questions like how to change a tire, and nobody
maintains a huge database of rules. Like the brain, the Internet continues
to work even when a large number of nodes or links are broken.

I'm guessing Kurzweil will win the bet. If he does, it will be because IT
people look to biology for ideas - not for superficial analogies, but for
real guidance.